Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Perplexity Launches Personal Computer for Mac, Turning a Mac mini Into an Always-On AI Agent
- Thread starter MacRumors
- Start date
- Sort by reaction score
Mykbibby
macrumors 6502a
ThailandToo
macrumors 65816
I think the possibilities are great, just like when the Internet came out. Quick regulation would be incredibly beneficial to humanity.
ThailandToo
macrumors 65816
Amazing Iceman
macrumors 603
Or Politicians? Oh wait, they already forgot the meaning of "work".
Amazing Iceman
macrumors 603
You'll be very Perplexed when you find out all your files were deleted because the AI considered these useless.
All the better to prevent you from unsubscribing.
Sounds like something Apple could build into MacOS eventually. I guess this is where it's heading. I need something that will organize my downloads folder, that would be great.
I wonder how many things people actually have on their to-do lists that AI could to. AI is without a doubt super useful for many kinds of work, but for "personal" tasks I can't think of much. Would of course be fantastic if it could tidy up the house, dust the furniture and do the laundry, but I doubt an AI on a Mac Mini will be of much help for that. Do I want it to go to the gym or read the news (and the Macrumors forum) for me? Not really. 😉 It's kind of telling that "sorting a messy downloads folder" is one of their examples ...
Skwoodge
macrumors regular
I haven't found anything that even comes close. I've been able to create two Swift apps with zero coding knowledge and a few Chrome and Firefox extensions. Gemini and ChatGPT have spat out scripts that don't even run.
What does this mean? " It can create teams of agents across over 20 frontier models to complete tasks" Agents doing what? Deleting my read emails?
Every AI model is always asking you, "Do you want me to do blah blah blah?" It has no purpose, which is why it is always asking you what it should do.
Does anyone have any examples of what they would use this for? I can’t think of anything I would do with it, but I’m not terribly creative…
PBG4 Dude
macrumors 603
You could always use it to manage your investment portfolio 🤪
In seriousness though, it seems like a great way to trawl the web if you want to do an extended search on a topic and let the AI cull unimportant data and bullet point the important stuff with links to source data/sites.
Lots of people are trying autonomous coding. For personal use I could see this being a positive. For open source and commercial software though, this could be a nightmare. Some OSS projects have disabled pull requests because of the code slop they’re being buried by.
Lone Deranger
macrumors 68000
boswald
macrumors 68040
Haiku_Oezu
macrumors 6502a
$200 a month for the privilege of watching a chatbot delete all your emails while you scream STOP
Skwoodge
macrumors regular
A rough analogy is what if you were running an small business. You would give the IT jobs to the 'IT guy'. The 'HR' stuff to the 'HR guy'. The accounting stuff to the 'Numbers gal' . and the communications/media gal would arrange the press releases , ad buys, and general communication. You 'hire' folks to do work based on their skill set that better than the other folks in the group. Everybody is not completely interchangeable.
The notion here is that different "Frontier' models have increasingly specialized skill sets. That runs very counter to how some folks are treating these models. Some are on the path of only giving everything to the one all knowing , all power model that does everything better than anybody. The latter set of folks don't want to buy this.
On the other side, it isn't all that mainstream for a generic users to have paid for AI chat subscriptions to 2-4 different model services vendors. ( buy account at Claude , OpenAI , and Perplexity). If don't have multiple accounts and not paying Perplexity for max model then this app actually doesn't work.
Perplexity's model (also in the cloud) orchestrates dispatching the work to various models. Each one of these models may have different upload/download points. Different intermediate files. Settings. etc. The Mini is basically a nexus point between a wide variety of vendors. ( and could keep some crticial files that are input/output local and out of the long term storage in the cloud. )
Over time I suspect some local models make take some of the workload. For example, perhaps tasks that deal with requesting local Mac app data or general file access go through Apple's AI agent that users trust more.
OpenAI dumping Sora is somewhat illustrative that the idea of the 'all knowing all seeing " model is dubious at least and possibly just smoke and a myth. Making the models bigger and bigger at some point likely is going to hit a wall. So if stop trying to be an expert at EVERYTHING then perhaps better to become differentially better at specific skills.
And if go back to context where have different workers for different jobs. There is common vocabulary that is used to coordinate with each other but each individual knows words and concepts that the others don't.
However, as others have talked about, the security model and protections here are somewhat vague. Also likely have 'holes' if this was all whipped together in a great hurry. ( e.g, AI tool whipped up most of it and no one has throughly testing it. ) All the orchestratin work is going up to Perplexity's clound and back down again. (possibly watch how all the tools are composed and likely has access to all of the shared files repository. )
That is a feedback gathering mechanism ( partial training data ) as much no purpose. There is limited memory and or reasoning context for most of these models. In a long drawn out dialog (if the model doesn't go spastic/hallucinatory ) on repeatitive tasks over and over again, the follow up questions pretty likely will get less 'random'.
I stop talking to it. It tries to bring up my job in almost ever context. "As a ____, you're probably going to be interested in the analytical.....blah blah."